refactor: Optimize bech32::Encode #13632

This is my treatment of @kallewoof's #13586

I got another big win out of not catting together the values and checksum for iteration, and a small win from using pre-increment.

This results in a ~40% speed-up in calls to Encode()

Before:

  #Benchmark, evals, iterations, total, min, max, median
  Bech32Encode, 5, 800000, 5.98941, 1.46948e-06, 1.56082e-06, 1.49065e-06

After:

  #Benchmark, evals, iterations, total, min, max, median
  Bech32Encode, 5, 800000, 3.60337, 8.82574e-07, 9.23948e-07, 8.93568e-07

Also throwing a std::length_error on length error, rather than asserting, for easier error handling.

This PR brought to you by insomnia.

You could cherry pick my benchmark as a separate commit since you seem to be using it as is (189cf35f3e6d2cc9ed08eb23dd0ea36be28b6c11).

Keeping bench and speed-up as separate commits is useful as you can compare easier.

@kallewoof Could you restore the branch? As of now it's on github but comes up as "fatal: bad object 189cf35f3e6d2cc9ed08eb23dd0ea36be28b6c11" via git.

Done

<!--e57a25ab6845829454e8d69fc972939a-->Note to reviewers: This pull request conflicts with the following ones:

• #13249 (Make objects in range declarations immutable by default. Avoid unnecessary copying of objects in range declarations. by practicalswift)

If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first.

Closing as per IRC comments that the current implementation "was made for simplicity and readability", "is not on the critical path", etc.